Routine Microsecond Molecular Dynamics Simulations with AMBER on GPUs. 1. Generalized Born

摘要

We present an implementation of generalized Born implicitsolventall-atom classical molecular dynamics (MD) within the AMBER programpackage that runs entirely on CUDA enabled NVIDIA graphics processingunits (GPUs). We discuss the algorithms that are used to exploit theprocessing power of the GPUs and show the performance that can beachieved in comparison to simulations on conventional CPU clusters.The implementation supports three different precision models in whichthe contributions to the forces are calculated in single precisionfloating point arithmetic but accumulated in double precision (SPDP),or everything is computed in single precision (SPSP) or double precision(DPDP). In addition to performance, we have focused on understandingthe implications of the different precision models on the outcomeof implicit solvent MD simulations. We show results for a range oftests including the accuracy of single point force evaluations andenergy conservation as well as structural properties pertaininingto protein dynamics. The numerical noise due to rounding errors withinthe SPSP precision model is sufficiently large to lead to an accumulationof errors which can result in unphysical trajectories for long timescale simulations. We recommend the use of the mixed-precision SPDPmodel since the numerical results obtained are comparable with thoseof the full double precision DPDP model and the reference double precisionCPU implementation but at significantly reduced computational cost.Our implementation provides performance for GB simulations on a singledesktop that is on par with, and in some cases exceeds, that of traditionalsupercomputers.